Facial Emotion Recognition Using Convolutional Neural Network Based on the Visual Geometry Group-19

Abstract

Facial recognition is one of many popular and difficult tasks in computer vision. A variety of research have been conducted on this subject, each of which suggests a stand-alone approach. While many studies strive for more accuracy, this study research aims to increase the efficiency of human computers by classifying emotions based on human faces using a self-based neural network. The usage of a Convolutional Neural Network (CNN) based on the Visual Geometry Group - 19 (VGG-19) classification model, which has been employed in ImageNet data sets and improved for emotion classification, is proposed in this paper. The classification process was conducted using FER-2013 dataset, which consists of over 35,000 facial images captured in various settings and contains 7 different emotions. The dataset was divided into three subsets, with 80% allocated for training, 10% for validation, and 10% for testing. With an accuracy of 71.80%, the proposed technique surpasses most self-based models.

Optimizing Facial Expression Recognition with Image Augmentation Techniques: VGG19 Approach on FERC Dataset

“…Table 4. Accuracy and Loss Method Accuracy (Prabaswera & Soeparno, 2023) 71.80% VGG19 (Without Augmentation) 62.73% VGG19 (Augmentation)…”

“…One of them is that the image quality in FERC datasets is often lower than the image quality obtained from digital cameras, which can affect the effectiveness of the algorithm in identifying faces (Febrian, Halim, Christina, Ramdhan, & Chowanda, 2022). The large number of FERC datasets also causes expensive computational costs and takes a lot of time (Prabaswera & Soeparno, 2023). The variability in faces such as differences in shape, size, pose, expression, lighting, and occlusion, also affects the ability of the model to generalize well (He, 2023).…”